Merge pull request #85 from Datura-ai/Update-Bittensor

add capacities

.gitignore

@@ -13,4 +13,8 @@ wandb/
 validators/.ipynb_checkpoints/
 **/validator.ipynb
 **/.env
-**/Cortex.t.egg-info
+**/Cortex.t.egg-info
+**/test.ipynb
+.env
+**/server/**/*.py
+cache.db

README.md

@@ -200,6 +200,7 @@ Before starting make sure update your system and have pm2 installed to run the s
 
 ```bash
 apt update -y && apt-get install git -y && apt install python3-pip -y
+
 ```
 
 Download the repository, navigate to the folder and then create virtual env and install the necessary requirements with the following chained command.
@@ -217,9 +218,11 @@ After installing it, copy `env.example` to `.env` and substitute
 all env vars with values appropriate for your accounts.
 
 ## Mining
-
+# step1.
+go to cortext/constants.py and change bandwidth_to_model value as per limit of api.
+currently we support only 3 models: "gpt-4o", "claude-3-5-sonnet-20240620", "llama-3.1-70b-versatile". 
+so don't add more than that.
 You can launch your miners via python3 using the following command.
-
 ```bash
 bash start_miner.sh
 ```

cortext/__init__.py

@@ -19,7 +19,7 @@
 
 
 # version must stay on line 22
-__version__ = "4.0.10"
+__version__ = "4.0.11"
 version_split = __version__.split(".")
 __spec_version__ = (
         (1000 * int(version_split[0]))
@@ -52,8 +52,6 @@
 IMAGE_BLACKLIST_STAKE = 5000
 EMBEDDING_BLACKLIST_STAKE = 5000
 ISALIVE_BLACKLIST_STAKE = min(PROMPT_BLACKLIST_STAKE, IMAGE_BLACKLIST_STAKE, EMBEDDING_BLACKLIST_STAKE)
-MIN_REQUEST_PERIOD = 2
-MAX_REQUESTS = 20
 # must have the test_key whitelisted to avoid a global blacklist
 testnet_key = ["5EhEZN6soubtKJm8RN7ANx9FGZ2JezxBUFxr45cdsHtDp3Uk"]
 test_key = ["5DcRHcCwD33YsHfj4PX5j2evWLniR1wSWeNmpf5RXaspQT6t"]
@@ -3769,3 +3767,6 @@
 
 
 ALL_SYNAPSE_TYPE = Union[StreamPrompting, Embeddings, ImageResponse, IsAlive]
+
+REDIS_RESULT_STREAM = 'result_stream'
+REDIS_RESULT = 'result'

cortext/constants.py

@@ -0,0 +1,87 @@
+TEXT_MODEL = "gpt-4-turbo-2024-04-09"
+TEXT_PROVIDER = "OpenAI"
+TEXT_MAX_TOKENS = 4096
+TEXT_TEMPERATURE = 0.001
+TEXT_WEIGHT = 1
+TEXT_TOP_P = 0.01
+TEXT_TOP_K = 1
+VISION_MODELS = ["gpt-4o", "claude-3-opus-20240229", "anthropic.claude-3-sonnet-20240229-v1:0",
+                 "claude-3-5-sonnet-20240620"]
+TEXT_VALI_MODELS_WEIGHTS = {
+    # from https://openai.com/api/pricing/
+    "OpenAI": {
+        "gpt-4o": 15.00,
+        # "gpt-3.5-turbo": 2.00,
+        # "o1-preview": 60.00,
+        # "o1-mini": 12.00,
+    },
+    # from https://ai.google.dev/pricing
+    # "Gemini": {
+    #     "gemini-1.5-flash": 0.30,
+    #     "gemini-1.5-pro": 10.50,
+    # },
+    #
+    "Anthropic": {
+        "claude-3-5-sonnet-20240620": 15.00,
+        # "claude-3-opus-20240229": 75,
+        # "claude-3-haiku-20240307": 1.25,
+    },
+    # model IDs from https://console.groq.com/docs/tool-use?hss_channel=tw-842860575289819136
+    # prices not available yet, default to bedrock pricing
+    # free tier: 30 rpm
+    "Groq": {
+        # "gemma2-9b-it": 0.22,
+        # "llama-3.1-8b-instant": 0.22,
+        "llama-3.1-70b-versatile": .99,
+        # "llama-3.1-405b-reasoning": 16,
+        # "mixtral-8x7b-32768": 0.7,
+    },
+    # from https://aws.amazon.com/bedrock/pricing/
+    # model IDs from https://docs.aws.amazon.com/bedrock/latest/userguide/model-ids.html#model-ids-arns
+    "Bedrock": {
+        # "mistral.mixtral-8x7b-instruct-v0:1": 0.7,
+        # "mistral.mistral-large-2402-v1:0": 24,
+        # "meta.llama3-1-8b-instruct-v1:0": 0.22,
+        # "meta.llama3-1-70b-instruct-v1:0": 0.99,
+        # "meta.llama3-1-405b-instruct-v1:0": 16,
+    }
+}
+
+bandwidth_to_model = {
+    "OpenAI": {
+        "gpt-4o": 2,
+        # "gpt-3.5-turbo": 1,
+        # "o1-preview": 1,
+        # "o1-mini": 1,
+    },
+    # from https://ai.google.dev/pricing
+    # "Gemini": {
+    #     "gemini-1.5-flash": 1,
+    #     "gemini-1.5-pro": 1,
+    # },
+    #
+    "Anthropic": {
+        "claude-3-5-sonnet-20240620": 2,
+        # "claude-3-opus-20240229": 1,
+        # "claude-3-haiku-20240307": 1,
+    },
+    # model IDs from https://console.groq.com/docs/tool-use?hss_channel=tw-842860575289819136
+    # prices not available yet, default to bedrock pricing
+    # free tier: 30 rpm
+    "Groq": {
+        # "gemma2-9b-it": 1,
+        # "llama-3.1-8b-instant": 1,
+        "llama-3.1-70b-versatile": 1,
+        # "llama-3.1-405b-reasoning": 16,
+        # "mixtral-8x7b-32768": 1,
+    },
+    # from https://aws.amazon.com/bedrock/pricing/
+    # model IDs from https://docs.aws.amazon.com/bedrock/latest/userguide/model-ids.html#model-ids-arns
+    # "Bedrock": {
+    #     "mistral.mixtral-8x7b-instruct-v0:1": 1,
+    #     "mistral.mistral-large-2402-v1:0": 1,
+    #     "meta.llama3-1-8b-instruct-v1:0": 1,
+    #     "meta.llama3-1-70b-instruct-v1:0": 1,
+        # "meta.llama3-1-405b-instruct-v1:0": 16,
+    # }
+}

cortext/protocol.py

@@ -15,7 +15,8 @@ class IsAlive(bt.Synapse):
 
 
 class Bandwidth(bt.Synapse):
-    bandwidth_rpm: Optional[Dict[str, int]] = None
+    bandwidth_rpm: Optional[Dict[str, dict]] = None
+
 
 class ImageResponse(bt.Synapse):
     """ A class to represent the response for an image-related request. """
@@ -124,6 +125,15 @@ class ImageResponse(bt.Synapse):
         description="A list of fields required for the hash."
     )
 
+    process_time: int = pydantic.Field(
+        default=9999,
+        title="process time",
+        description="processed time of querying dendrite.",
+    )
+    task_id: str = pydantic.Field(
+        default="9999"
+    )
+
     def deserialize(self) -> Optional[Dict]:
         """ Deserialize the completion data of the image response. """
         return self.completion
@@ -286,6 +296,14 @@ class StreamPrompting(bt.StreamingSynapse):
         title="streaming",
         description="whether to stream the output",
     )
+    deserialize_flag: bool = pydantic.Field(
+        default=True
+    )
+    task_id: str = pydantic.Field(
+        default="9999",
+        title="task_id",
+        description="task id of the request from this syanpse."
+    )
 
     async def process_streaming_response(self, response: StreamingResponse) -> AsyncIterator[str]:
         if self.completion is None:
@@ -297,9 +315,6 @@ async def process_streaming_response(self, response: StreamingResponse) -> Async
                     self.completion += token
             yield tokens
 
-    def deserialize(self) -> str:
-        return self.completion
-
     def extract_response_json(self, response: StreamingResponse) -> dict:
         headers = {
             k.decode("utf-8"): v.decode("utf-8")
@@ -332,4 +347,4 @@ def extract_info(prefix: str) -> dict[str, str]:
             "timeout": self.timeout,
             "streaming": self.streaming,
             "uid": self.uid,
-        }
+        }

cortext/reward.py

@@ -18,8 +18,8 @@
 # DEALINGS IN THE SOFTWARE.
 from __future__ import annotations
 from transformers import logging as hf_logging
-hf_logging.set_verbosity_error()
 
+hf_logging.set_verbosity_error()
 
 import re
 import io
@@ -37,10 +37,13 @@
 from sklearn.feature_extraction.text import TfidfVectorizer
 from transformers import CLIPProcessor, CLIPModel
 
+
 # ==== TEXT ====
 
 def calculate_text_similarity(text1: str, text2: str):
     try:
+        text1 = str(text1).lower()
+        text2 = str(text2).lower()
         # Initialize the TF-IDF Vectorizer
         vectorizer = TfidfVectorizer()
 
@@ -50,12 +53,12 @@ def calculate_text_similarity(text1: str, text2: str):
         # Calculate the Cosine Similarity
         similarity = cosine_similarity(tfidf_matrix[0:1], tfidf_matrix[1:2])[0][0]
 
-        bt.logging.debug(f"Similarity: {similarity}")
         return similarity
     except Exception as e:
         bt.logging.error(f"Error in calculate_text_similarity: {traceback.format_exc()}")
         raise
 
+
 async def api_score(api_answer: str, response: str, weight: float, temperature: float, provider: str) -> float:
     try:
         if api_answer is None or response is None:
@@ -153,6 +156,7 @@ def calculate_image_similarity(image, description, max_length: int = 77):
     # Calculate cosine similarity
     return torch.cosine_similarity(image_embedding, text_embedding, dim=1).item()
 
+
 async def dalle_score(uid, url, desired_size, description, weight, similarity_threshold=0.21) -> float:
     """Calculate the image score based on similarity and size asynchronously."""
 
@@ -191,11 +195,11 @@ async def dalle_score(uid, url, desired_size, description, weight, similarity_th
         return 0
 
 
-
 # IMAGES ---- DETERMINISTIC
 
 async def deterministic_score(uid: int, syn, weight: float):
-    vali_b64s = await utils.call_stability(syn.messages, syn.seed, syn.steps, syn.cfg_scale, syn.width, syn.height, syn.samples, syn.sampler)
+    vali_b64s = await utils.call_stability(syn.messages, syn.seed, syn.steps, syn.cfg_scale, syn.width, syn.height,
+                                           syn.samples, syn.sampler)
 
     for miner_b64, vali_b64 in zip(syn.completion["b64s"], vali_b64s):
         if miner_b64[:50] != vali_b64[:50]:
@@ -206,15 +210,13 @@ async def deterministic_score(uid: int, syn, weight: float):
     return weight
 
 
-
 # ==== Embeddings =====
 
 async def embeddings_score(openai_answer: list, response: list, weight: float, threshold: float = .95) -> float:
     if len(openai_answer) != len(response):
         bt.logging.info("The number of embeddings in openai_answer and response do not match.")
         return 0
 
-
     # Calculate similarity for each pair of embeddings
     similarities = []
     for oa_emb, resp_emb in zip(openai_answer, response):

cortext/utils.py

@@ -310,7 +310,7 @@ async def get_item_from_list(items, vision):
     async with list_update_lock:
         items = state[category][item_type]
 
-        bt.logging.debug(f"Queue for {list_type}: {len(items) if items else 0} items")
+        bt.logging.trace(f"Queue for {list_type}: {len(items) if items else 0} items")
 
         item = await get_item_from_list(items, vision)
 
@@ -319,7 +319,7 @@ async def get_item_from_list(items, vision):
             items = await get_items(category, item_type, theme)
             bt.logging.trace(f"Items generated: {items}")
             state[category][item_type] = items
-            bt.logging.debug(f"Fetched new list for {list_type}, containing {len(items)} items")
+            bt.logging.trace(f"Fetched new list for {list_type}, containing {len(items)} items")
 
             item = await get_item_from_list(items, vision)
 
@@ -457,8 +457,8 @@ def extract_python_list(text: str):
 
 async def call_openai(messages, temperature, model, seed=1234, max_tokens=2048, top_p=1):
     for _ in range(2):
-        bt.logging.debug(
-            f"Calling Openai. Temperature = {temperature}, Model = {model}, Seed = {seed},  Messages = {messages}"
+        bt.logging.trace(
+            f"Calling Openai to get answer. Temperature = {temperature}, Model = {model}, Seed = {seed},  Messages = {messages}"
         )
         try:
             message = messages[0]
@@ -503,7 +503,7 @@ async def call_openai(messages, temperature, model, seed=1234, max_tokens=2048,
 
 
 async def call_gemini(messages, temperature, model, max_tokens, top_p, top_k):
-    bt.logging.debug(f"Calling Gemini. Temperature = {temperature}, Model = {model}, Messages = {messages}")
+    bt.logging.trace(f"Calling Gemini. Temperature = {temperature}, Model = {model}, Messages = {messages}")
     try:
         model = genai.GenerativeModel(model)
         response = model.generate_content(
@@ -554,7 +554,7 @@ async def call_gemini(messages, temperature, model, max_tokens, top_p, top_k):
 
 async def call_anthropic_bedrock(prompt, temperature, model, max_tokens=2048, top_p=1, top_k=10000):
     try:
-        bt.logging.debug(
+        bt.logging.trace(
             f"Calling Bedrock via Anthropic. Model = {model}, Prompt = {prompt}, Temperature = {temperature}, Max Tokens = {max_tokens}"
         )
         completion = await anthropic_bedrock_client.completions.create(
@@ -610,7 +610,7 @@ async def generate_messages_to_claude(messages):
 
 async def call_anthropic(messages, temperature, model, max_tokens, top_p, top_k):
     try:
-        bt.logging.info(
+        bt.logging.trace(
             f"calling Anthropic for {messages} with temperature: {temperature}, model: {model}, max_tokens: {max_tokens}, top_p: {top_p}, top_k: {top_k}"
         )
         filtered_messages, system_prompt = await generate_messages_to_claude(messages)
@@ -633,7 +633,7 @@ async def call_anthropic(messages, temperature, model, max_tokens, top_p, top_k)
 
 async def call_groq(messages, temperature, model, max_tokens, top_p, seed):
     try:
-        bt.logging.info(
+        bt.logging.trace(
             f"calling groq for {messages} with temperature: {temperature}, model: {model}, max_tokens: {max_tokens}, top_p: {top_p}"
         )
 
@@ -655,7 +655,7 @@ async def call_groq(messages, temperature, model, max_tokens, top_p, seed):
 
 async def call_bedrock(messages, temperature, model, max_tokens, top_p, seed):
     try:
-        bt.logging.info(
+        bt.logging.trace(
             f"calling AWS Bedrock for {messages} with temperature: {temperature}, model: {model}, max_tokens: {max_tokens}, top_p: {top_p}"
         )
 
@@ -746,7 +746,7 @@ async def extract_message(message):
 
 async def call_stability(prompt, seed, steps, cfg_scale, width, height, samples, sampler):
     # bt.logging.info(f"calling stability for {prompt, seed, steps, cfg_scale, width, height, samples, sampler}")
-    bt.logging.info(f"calling stability for {prompt[:50]}...")
+    bt.logging.trace(f"calling stability for {prompt[:50]}...")
 
     # Run the synchronous stability_api.generate function in a separate thread
     meta = await asyncio.to_thread(

miner/config.py

@@ -36,7 +36,8 @@ def __init__(self):
 
         self.BT_SUBTENSOR_NETWORK = 'test' if self.ENV == 'test' else 'finney'
         self.WANDB_OFF = False if self.ENV == 'prod' else True
-        self.LOGGING_TRACE = False if self.ENV == 'prod' else True
+        # still can use the --logging.debug and --logging.trace to turn on logging
+        self.LOGGING_TRACE = False # if self.ENV == 'prod' else True
         self.BLACKLIST_AMT = 5000 if self.ENV == 'prod' else 0
         self.BLOCKS_PER_EPOCH = int(os.getenv('BLOCKS_PER_EPOCH', 100))
         self.WAIT_NEXT_BLOCK_TIME = int(os.getenv('WAIT_NEXT_BLOCK_TIME', 1))